A simple application-level proxy for the pathfinder's RPC protocol.
The vanilla pathfinder listens on a configurable port for incoming rpc-calls. Some rpc functions allow to change the pathfinder's state which should not be possible from outside. This proxy limits access to only whitelisted rpc-functions and also acts as a load balancer and auto scaler.
The proxy is still in development and not yet ready for production.
The proxy can be configured with the following environment variables:
(Note: every upstream must have a corresponding health check entry)
The port the proxy listens on.
A comma separated list of allowed origins for CORS:
https://my-first-client-app.com,https://my-second-client-app.com
A comma separated list of URLs that point to upstream pathfinder instances:
https://pathfinder-1.com,https://pathfinder-2.com
A comma separated list of URLs that point to the health endpoints of the upstream pathfinders:
https://pathfinder-1.com/health,https://pathfinder-2.com/health
- Clone the repository
- Install dependencies with
npm install - Compile typescript with
npx tsc - Start the proxy with
node dist/main.js
# Get the code
git clone https://github.com/CirclesUBI/pathfinder-proxy.git
# Build
cd pathfinder-proxy
npm install
npx tsc
# Configure
PORT=5551
CORS_ORIGINS=https://my-first-client-app.com,https://my-second-client-app.com
UPSTREAM_SERVICE_ENDPOINTS=https://pathfinder-1.com,https://pathfinder-2.com
UPSTREAM_HEALTH_ENDPOINTS=https://pathfinder-1.com/health,https://pathfinder-2.com/health
# Run
node dist/main.jsThe proxy listens on port 4999 for incoming requests.
It accepts POST requests to the / path.
You can use the following curl command to test the proxy:
curl --location --request POST 'localhost:4999' \
--header 'Content-Type: application/json' \
--data-raw '{
"id":"875634785",
"method": "compute_transfer",
"params": {
"from": "0x052b4793d50d37FD3BFcBf93AAC9Cda6292F81Fa",
"to": "0x42cEDde51198D1773590311E2A340DC06B24cB37",
"value": "9999999999999999999999999999"
}
}'The proxy maintains a list of upstreams. Each upstream is a pathfinder2 instance and has an associated health status. Unhealthy instances will not receive any calls.
The proxy performs a health check on every upstream and blocks access as long as the upstream is unhealthy. Health checks are just http requests. The proxy expects a 200 status code for a healthy upstream service and 500 for an unhealthy one.
Associated with every upstream is a set of statistics. These are updated on every call. Currently only the average response time is tracked for 30, 60 and 120 second windows.
The proxy maintains a whitelist of rpc-functions. Only calls to whitelisted functions are forwarded to an upstream. The filters compare the 'method' field of the rpc-call against the whitelist.
The only externally available method for now is "compute_transfer".
The requests are distributed to the upstreams using round-robin.
Based on the statistics, the proxy must be able to decide if new cluster-nodes should be added or if some can be removed. It should use the kubernetes api from within the pod to scale the pathfinder deployment.